Possible RAM problems might be causing system instability.
Possible RAM problems might be causing system instability.
Well, things have definitely improved in some areas since my last time working on my computer, so I'm excited to give it another try.
You can also adjust all other timing settings by making small changes, though I’m not confident enough to give specific instructions. There are a few people here who understand RAM tuning well; you might ask them in the memory sub thread for advice. I don’t always tag you for this topic, but you’re definitely the most knowledgeable person I know about it. @RONOTHAN## is great, and @Eigenvektor is also smart. You seem to have a good grasp of the subject.
Na, I don't really mind at all. OP, if you really want to run 4x16GB, there's a couple things you should try to do. First things first is to make sure that you're sticks are all actually the same, since if you're running a RAM config with mixed memory dies, tuning this will be an absolute nightmare and it's just not worth the migraine. There's a couple ways of doing that, though the easiest is to just figure out the max frequency of each stick. Set the memory to voltages to 1.4V and the primary timings to 40-50-50-128, then start increasing the memory frequency until it stops booting (this will take quite a while to boot, so be prepared for 2-3 minutes per boot). If it does 6400 at the max, it's likely Samsung, if it does ~6800-7000 at the max it's likely Hynix 16Gb M die, and if it does 7600+ it's Hynix 16Gb A die. There are other ways to figure it out by looking at the label (with Kingston kits, this can only confirm if its Samsung, not the Hynix revision), checking tRFC (this more works on Intel/APUs though for complicated reasons), or using a program called Thaiphoon Burner (this doesn't always work and can give some weird issues on DDR5/Windows 11, so keep that in mind), so going for the max frequency is probably the most consistent option, albeit the most time consuming. After you figure out if everything is actually the same die revision, try messing around with a couple different BIOS revisions. Over the past ~year, AMD has more or less alternated between AGESA versions that have had 4x16GB working well and being completely broken. Newer does not always mean better for the BIOS, and if you find the right BIOS revision you might be able to get even 6000+ to work with 4 DIMMs. I should probably mention that the method I'm giving more or less ignores the XMP/EXPO profile on those DIMMs as almost any kit rated at 5600 CL36 should do at least DDR5 6000, so you might as well get that extra frequency to work if you're gonna have to tune this slightly anyway Once you have the BIOS that you're running, it is somewhat time to get the max frequency to work with 4 DIMMs. Keep using those timings from before (you can lower them later), then start raising the memory frequency until you have instability in a stress test called Y-Cruncher VT3 (this is arguably the best test for finding memory instability due to high frequency). Auto settings should get up until 6000 to work, though once you go to 6200 you might need to mess with the SOC voltage (higher is better for maximum frequency, though don't go above 1.3V) and you'll need to manually set it to MEMCLK == UCLK. After the max frequency is discovered, it's time to start lowering the primary timings until you have stability issues. These will vary greatly on what memory chip you have, though if you have one of the Hynix variance, it'll be something between 30-36-32-30 and 32-40-40-40 and if you have Samsung it'll almost certainly be 36-36-36-76. If you want to just figure it out yourself, lower them one at a time until it starts crashing in memory stress tests (TestMem5 is a good free one, though admittedly it's a little hard to use). You can then try tuning the sub-timings afterwards, as those have a quite a bit of a performance uplift. Those you can find manually, though it'll probably be best to just find a Buildzoid timings video with your specific IC and just copy the timings from it. LMK if you have any questions, almost certainly something in there was confusing and I'll need to clarify it.
I understand your concern. I can handle general questions about RAM, though I’m limited on specific tuning details, particularly regarding DDR5. You’re in the right place!
You spent some time tinkering after work and managed to improve performance. You changed the chipset to DDR5-5400 and adjusted settings to match the EXPO profile for DDR5-5600 (values 36-38-38-80-118-85-827-449-365). Interestingly, the tRC setting didn’t accept your input; it stayed at 85 even when you changed it. The memory modules look identical except for serial numbers and manufacturing months—one batch is from week 43 of 2023, the other from week 51. Running TestMem5 after 30 minutes showed no errors, though the 5600 MHz frequency had issues in the first minute, sometimes requiring a full reset. You also couldn’t resolve the Y-Cruncher test; it seemed to stop after reaching 44GB, possibly due to decimal limits or an internal error.
The default setting of 85 likely refers to tWR, not tRC. On AMD processors, WR can only be adjusted in multiples of six, which explains why it settled at 84. The correct value should be 118, matching tRP plus tRAS. The full configuration sequence is tCL-tRCD-tRP-tRAS-tRC-tWR-tRFC-p. There are three SK Hynix variants: 16Gb A, 16Gb M, and 24Gb M. You probably have a 16Gb model, though confirming requires checking the maximum boot frequency. The 16Gb M die usually handles over 6800 MHz on AMD, while the A die can reach 8000+ if supported. I’d guess both are 16Gb M, but testing is wise. Y-Cruncher feels odd to operate; I should have clarified its usage. You can access it via GUI by launching the .exe, selecting stress mode, running VT3, and hitting clear tests. For automation, use CMD: navigate to the directory, run “y-cruncher stress VT3”, and let it execute for two hours. You can also create a batch file with a time limit like “y-cruncher stress VT3 -TL:7200”.
Finally took some time to experiment a bit more, man this week’s been crazy... anyways, I ran the stress tests and so far here’s what I found: DDR5-5400 CL36 (manually adjusted): Y-Cruncher triggered an error after run 9 TestMem5. DDR5-5200 CL36 (manually adjusted): Y-Cruncher failed on run 46 TestMem5, with an error after about 1 hour and 35 minutes. I attached a few screenshots of the exact errors since they look different. Also, you were right—I mistakenly set the threshold at 85 on tWR, but I was correct when I noted it should be 85 on EXPO. I included a picture of the preset I used to manually configure timings (EXPO 0). I’d been using DDR5-5400 for a few days without issues, but after a Skype call yesterday I unplugged a USB mic from my PC and the whole system restarted immediately. Not sure if this is related to RAM or something new, so I’ll keep lowering the settings. Probably won’t fix anything significant, but I also sent an email to ASRock today asking for advice on recommended settings (maybe someone else has faced the same problem).
It seems crashes in Y-Cruncher and TM5 point toward memory-related problems. If your system was stable before, it should continue running smoothly at any desired memory speed. I’m not questioning this, but Intel’s XMP profiles store timing data, and while 85 works on some systems, it must be adjusted to 84 for AMD. Hynix memory typically hits 48 on AM5 platforms, so rounding down a bit from the EXPO is acceptable for stability. You might want to verify other BIOS updates, as some AM5 versions are less reliable with four DIMMs and others perform similarly to two 32GB units. Your setup appears to be leaning toward the less stable side.
Will likely continue testing over the next few days to achieve maximum stability. Thanks to everyone in this thread—gained a lot of new insights I hadn’t considered before! ^-^ I made an edit for those who might ask the same question: tried four different BIOS updates starting from version 2.10, but all produced similar outcomes. Returning to 2.10 and conducting more tests showed that only DDR5-3800 with manual CL36 timings remained fully stable. I reached out to ASRock’s support, who stated that four sticks will work only if they’re part of a complete four-stick package, not just two separate kits. Since testing such low frequencies feels impactful, I decided to return the RAM set and save the money for something else.